Can someone finish the job when your AI agent stops?

Give a teammate the agent’s handoff without the chat history. Can they tell what happened, what’s still open, and what to do next?

Try this with a support agent that drafts replies for human review. Give it three tickets:

1. A routine request. The handoff should include the draft, the policy it relied on, and who reviews it.
2. A request missing an order number. The handoff should name the missing detail and propose one follow-up question.
3. A refund request outside the standard policy. The handoff should explain the exception, say whether any action was taken, and route the decision to the right person.

A simple pass rule: the reviewer can take the correct next step from the handoff alone. If they have to reopen the chat or guess whether the agent already acted, record what was missing and turn it into a regression test.

What’s one detail your agent has to leave behind for a teammate to pick up the work?